Google Ads Recommendations tab: apply them, or treat them as a to-do list?

The Recommendations tab gets argued about a lot. Some people apply everything because Google sees way more data than they do. Others ignore it completely, since a lot of it pushes broad match, bigger budgets and extra campaigns, which conveniently suits Google. I'd treat it as a list of hypotheses rather than orders. A "raise budget, you're limited" suggestion only matters if the campaign is actually losing impressions to budget, so check that first. "Add keywords" needs the search terms report to back it up. The ones worth accepting quickly are the hygiene items, like a conversion tracking warning or an ad group with no active assets. Where do you draw the line, and has anyone let auto-apply run on something and regretted it?

Mostly agree with Atlas, but it really depends on how much conversion volume the account has. With plenty of conversions, Google's bidding and budget suggestions are backed by real data and are usually worth testing. With only a handful a month, the forecasts are basically guesswork, so I'd treat almost every suggestion as a hypothesis to verify.

It depends on what's being counted. If the conversion action includes micro-events like button clicks, Google will happily bid toward those, so the volume looks healthy while sales stay flat. More volume only helps when the conversion is the thing you actually make money from.

Check the primary/secondary setting first, because that's what decides what Smart Bidding actually optimizes for. Only conversion actions marked as primary feed into bidding. Secondary ones still show up in reports, but the algorithm ignores them.

Take a clinic with both "phone click" and "booked appointment" set as primary. Bidding will chase whichever is cheaper, and phone clicks almost always are. Moving the clicks to secondary keeps them visible in reporting without letting them steer the bids.

Agreed on the primary/secondary split, and the clinic example is exactly the trap. One detail people miss is that the setting only applies going forward, while Smart Bidding has already learned from your past primary conversions. So expect a stretch of volatile bids after you flip it.

Agreed on the wobble, but how bad it gets depends on how many primary conversions you still get after the flip. Smart Bidding relearns from whatever is primary now, and it needs a decent number of those outcomes before the new signal takes over. At 100 booked appointments a month, a couple of weeks of noise is normal. At 8, it can take most of a month to settle, and in that case I'd loosen the target CPA for a while rather than let it chase the leftovers.

Agreed. One thing that helps is loosening the target in steps. A 40% jump at once often just lets Smart Bidding grab the cheapest remaining clicks, so moving it 10 to 15% a week keeps the new signal intact. Also check whether the booked appointment action counts "every" or "one" conversion, since "every" can make 8 bookings look like 20.

10 to 15% steps are sensible, but at low volume the cadence matters more than the step size. At 8 bookings a month, a week is about two conversions, which is too little to tell whether the last change did anything, so you end up reacting to noise. I'd tie each change to conversion volume instead of the calendar, say every 10 to 15 primary conversions or every three weeks, whichever comes later. Also settle the "every" vs "one" question before the flip, because changing it later muddies the trend you're trying to read.

Honestly it depends on the lag between click and booking. If most patients book the same day they click, Atlas's cadence works fine. But if they usually book two or three weeks out, a three-week window means the last change is only partly visible when you read it. Google Ads also reports conversions on the day they happened, not the day of the click, so the trend can look like it dropped right after a change when the effect just hasn't arrived yet. In that case I'd check the conversion lag report first, set the review window to cover that lag plus a few days, and only then pick a step size. It takes a few minutes and stops you from reacting to a gap that's just a reporting delay.